kickc/manual/kickc.md

85 KiB
Raw Permalink Blame History

BETA

KickC is currently in beta, and occasionally crash with a cryptic error or create ASM code that does not work properly. Feel free to test it and report any problems or errors you encounter. Also, be prepared that breaking changes (to syntax, to semantics, etc.) may be implemented in the next versions.

1 What is KickC?

KickC is a C-compiler for 6502-based platforms creating optimized and readable assembler code.

The KickC language is classic C with a few limitations, and a few extensions to ensure an optimal fit for creating 6502 assembler code.

1.1 The KickC Language

The language is 95% standard C and supports most of the basic features of C, so it should be quite easy to get started with if you have programmed in C or any similar language.

Here is a simple “hello world” program, that prints “hello world!” on the screen.

#include <stdio.h>
void main() {
	printf(hello world!);
}

The language has some limitations compared to standard C, for example no support for unions or reentrant functions. Some features were omitted because they cannot be realized in a way that creates optimized 6502 assembler code. Others were omitted simply because they have not yet been implemented in the language.

The language also has a few extensions to standard C. The modifications and extensions were included either to allow creation of better 6502 assembler code or for convenience.

All limitations, modifications and extensions are described in the following sections.

1.2 Supported 6502-based platforms

The KickC-compiler includes all necessary linker and header files to makes it easy to create and test binaries for the following 6502-based platforms out-of-the-box:

  • Commodore VIC 20
  • Commodore 64
  • Commodore Plus/4 ( also usable for Commodore 16 and Commodore 116)
  • Atari 2600
  • Atari XL/XE
  • Nintendo NES
  • MEGA65
  • Commander X16

For convenience KickC supports compiling and running in a suitable emulator directly from the command line.

If you wish to develop for another 6502-based platform, then you can create your own linker and header-files for that platform.

KickC uses the very versatile Kick Assembler. The KickC Compiler produces assembler code for the MOS Technology 6502 processor family. Specifically the compiler supports 6502, 65C02, 65CE02 and 45GS02 CPUs.

1.3 Optimized and Readable Assembler Code

The compiler uses a number of modern optimization methods to create 6502-family assembler code that executes as fast as possible and does not contain unnecessary boilerplate. The optimization techniques include:

  • Detection of constant values and expressions
  • Optimized allocation of registers to variables
  • Optimized parameter and return value passing to/from functions
  • Minimizing the number of zero-page addresses used for storing variables
  • Choosing optimal assembler instructions to represent each statement
  • Removing unused functions, variables and code
  • Peephole optimization of the generated assembler code

The optimization techniques are also explained in more detail in later sections.

Below is a slightly more complex version of hello world, which prints “hello world!” with an added space between each letter on the first and third line of the C64 default screen at 0x400. This example illustrates how the KickC compiler creates optimized readable 6502 assembler.

helloworld2.c

char* screen = 0x400;

void main() {
    char* hello = "hello world!";
    print2(screen, hello);
    print2(screen+2*40, hello);
}

void print2(char* at, char* msg) {
    char j=0;
    for(char i=0; msg[i]; i++) {
        at[j] = msg[i];
        j += 2;
    }
}

helloworld2.asm

.label screen = $400
main: {
    lda #<screen
    sta print2.at
    lda #>screen
    sta print2.at+1
    jsr print2
    lda #<screen+2*$28
    sta print2.at
    lda #>screen+2*$28
    sta print2.at+1
    jsr print2
    rts
    hello: .text "hello world!@"
}
print2: {
    .label at = 2
    ldy #0
    ldx #0
  b1:
    lda main.hello,x
    sta (at),y
    iny
    iny
    inx
    lda main.hello,x
    cmp #0
    bne b1
    rts
}

The KickC compiler uses the following insights to optimize the helloworld2 program:

  • The screen pointer is never modified and is therefore a constant location in memory.
  • The second parameter to the print2-function (msg) always has the same value main.hello, so that can be hardcoded inside the method body instead of parsing it.
  • The contents of msg[i] (ie. the hardcoded main.hello string) can be addressed using simple indexing.
  • The at-parameter in print2 is truly variable, so it is placed on zero-page ($2 and $3) and indirect indexing can be used for addressing the contents of at[j].
  • The X-register is optimal for the i variable in the print2 function as it is good at indexing and incrementing.
  • The Y-register is optimal for the j variable in the print2 function as it is good at indirect indexing and incrementing.
  • When 2 is added to the j-variable it is better to do INY twice than to use addition.
  • The A-register is optimal for holding the current character c of the message being moved to the screen as LDA and STA can be efficiently indexed by X and indirect indexed by Y.

When generating Kick Assembler code the KickC compiler tries to ensure that the assembler code is readable and corresponds to the source C program as much as possible. This includes using the same names, using scopes and recreating constant calculations in assembler, when possible.

The KickC Compiler creates readable 6502 assembler code for the helloworld2 program by:

  • Using the variable and function-names screen, hello, main, print2, at in the generated code.
  • Recreating the calculation of the constant screen+2*40 in the assembler code as screen+2*$28.
  • Creating a local named scope in the assembler-code for the methods main and print2 by enclosing them in curly braces.
  • Placing method-local data and labels inside the method scope.This allows other assembler code to access the local data/labels using dot-syntax eg. main.hello or print2.at.

1.4 Getting Started

The KickC development is hosted on gitlab (https://gitlab.com/camelot/kickc). Here it is possible to follow the development and to download the latest binary release.

You install KickC on your own computer by:

The zip-file contains the following:

  • A bin Folder containing bat/sh-files for running KickC.

  • An examples Folder containing some example KickC programs.

  • An include Folder containing header-files for useful library functions usable in your own program.

  • A lib Folder containing C-files implementing the useful library functions usable in your own program.

  • A jar Folder containing the KickC JAR-file plus a few other JARs needed for running KickC (antlr4-runtime, picocli and KickAssembler).

  • A target Folder containing target platform specification and linker files for each supported platform.

  • A fragment Folder containing small asm-files used by the compiler to create the assembler code.

  • This manual in PDF-format and some files with license-information.

KickC is written in Java. To use KickC you need to download and install a Java runtime from https://www.java.com or https://jdk.java.net/. Java must be added to your PATH, or the environment variable JAVA_HOME must point to the folder containing the Java installation.

[!NOTE] KickC runs a lot faster on 64bit Java than on 32bit Java. You are therefore encouraged to ensure that your Java is a 64bit version. You can check by executing java -version in a Terminal/Command Prompt.

After installing KickC and you can compile a simple sample KickC program by doing the following:

MacOS

  • Start a Terminal
  • cd to the folder containing KickC
  • Enter the command bin/kickc.sh -e examples/helloworld/helloworld.c

Windows

  • Start a Command Prompt
  • cd to the folder containing KickC
  • Enter the command bin\kickc.bat -e examples\helloworld\helloworld.c

This compiles the helloworld KickC program examples/helloworld/helloworld.c
and produces assembler code in examples/helloworld/helloworld.asm.

The command line option -e instructs the compiler to both assemble the ASM code and execute the resulting binary program in a suitable emulator.

If you prefer not to assemble and run the code then just leave the -e option out. The compiler will then just produce an ASM-file that can then be assembled using KickAssembler, producing a runnable program that can be executed using an emulator or transfered to the real 8-bit hardware.

The emulator must of course be installed on your machine and available in the PATH for (-e) to work properly. KickC supports compiling for a number of different platforms, and uses different emulators for each platform. The following are the emulators used by the built-in target platforms:

To compile, assemble and execute the example Commodore 64 program simple-multiplexer.c (a sprite multiplexer moving 32 balloons in a sine on the screen) in the VICE emulator use the following command in the kickc-folder (assuming MacOs).

To create your own KickC-programs use any text-editor to create a source file containing a void main() function, save it to the file system (using the .c extension is recommended) and compile it by passing it to kickc.sh (on MacOs) or kickc.bat (on Windows).

2 KickC Language Reference

KickC is a C-family language, and 95% of the syntax and semantics is the same as C99. In the following the different parts of the language are explained. Finally the differences between KickC and standard C are listed.

2.1 Variables

The selected text discusses variable declaration in C programming language. Here are the key points:

  • Variables are declared by type name and can include an optional initialization assignment. For example, char size = 12; declares a signed char variable named 'size' with an initial value of 12.
  • If variables do not have an initial assignment, they will be initialized with the default value zero.
  • It's possible to declare multiple variables of the same type by separating the names and optional initializations with commas. For example, unsigned int a = 4, b; declares two unsigned int variables 'a' and 'b'. 'a' is initialized to 4 while 'b' is initialized to the default value (zero).
  • In KickC, variables can be declared at any point in the program, outside functions or inside function declarations.

2.2 Data Types

2.2.1 Integers

KickC supports the standard C integer data types such as char, short, int, long. Additionally, it introduces some fixed-size integer types that are more familiar on the 6502 platform: byte, word, dword.

Type Name Description
char, unsigned char, byte, unsigned byte An unsigned 8 bit (1 byte) integer. Range is [0;255].
signed char, signed byte A signed 8 bit (1 byte) integer in twos complement. Range is [-128;127].
unsigned short, unsigned int, unsigned, word, unsigned word An unsigned 16 bit (2 byte) integer. Range is [0;65,535].
short, signed short, int, signed int, signed, signed word A signed 16 bit (2 byte) integer in twos complement. Range is [32,767; +32,767].
unsigned long, dword, unsigned dword An unsigned 32 bit (4 byte) integer. Range is [0, 4,294,967,295].
long, signed long, signed dword A signed 32 bit (4 byte) integer in twos complement. Range is [2,147,483,647, 2,147,483,647].
  • If the 8-bit integer types (char, byte) are declared without an unsigned/signed prefix, they default to unsigned.
  • If the C 16+bit integer types (short, int, long) are declared without an unsigned/signed prefix, they default to signed, except char which defaults to unsigned.
  • If the 16+bit 6502-friendly integer types (word, dword) are declared without an unsigned/signed prefix, they default to unsigned.
  • Integer literals can be either decimal, hexadecimal or binary. The syntax for hexadecimal integer literals supports both C syntax (prefixing with 0x) and 6502 assembler syntax (prefixing with $). Similarly, the syntax for binary supports both prefixing with 0b and %.
Prefix Format Examples
Decimal No prefix 12, 53280
Hexadecimal 0x or $ 0x40, $dc01
Binary 0b or % 0b101, %1100110011001100

Character Literals: Character literals are unsigned bytes/chars. The syntax for a character literal is the character enclosed in single quotes. For example, char c = 'c';.

Escape Sequences: Escape sequences can be used to represent special characters such as newline. The following character escape sequences are supported:

  • '\n': newline
  • '\r': carriage return
  • '\f': form feed
  • '\': single quote
  • '\"': double quote
  • '\\': backslash

Character Encoding: Numerically, a character is represented by whatever value the character has in the current character encoding. The default encoding depends on the platform you are compiling for. You can change the encoding by changing the target platform or by using #pragma encoding().

2.2.1 Booleans

KickC also has a boolean type called bool. The boolean literals are true and false. Underlying the boolean type is a byte containing either 0 (if false) or 1 (if true). The following is an example of a boolean variable called enabled.

bool enabled = true;

2.2.2 Pointers

Pointers to all integer types and booleans are supported and declared using the syntax type*. The following is an example, where screen is a pointer to a char and pos is a pointer to a signed integer.

char* screen; 
int* pos;

Pointers to pointers are also supported. Here an example of a pointer to a pointer to an unsigned char.

unsigned char** screenptr = &screen; 

For functions it is only possible to use pointers to functions that has no return value and take no parameters.

2.2.3 Arrays

Arrays are supported using the syntax type a[] or type a[size]. For most practical purposes array variables are treated exactly like pointers.

Arrays can be initialized by an array literal written as comma-separated values inside curly braces eg. { 1, 2, 3 }. They can also be initialized with all zero values just by declaring the array to have a specific size.

String literals can also be used to initialize an array of unsigned bytes. The syntax for a string literal is a string enclosed in double quotes. Strings can use escape sequences to represent special characters such as newline. See character literals above for a list of the escape sequences. Stings are per default zero terminated. It is possible to create a string that is not zero terminated by adding the special suffix z after the last double quote.

Arrays that are initialized will allocate the memory needed for the size.

In the following example, sums is array of 3 signed chars initialized with zeros, fibs is an array of 6 signed integers containing the first Fibonacci numbers. msg is an unsigned char array containing the numeric value of the 5 characters h, e, l, l and o plus a sixth value that is zero (because strings are zero terminated) while msg2 only has the 5 characters without the final zero. Finally bs is simply a pointer to a boolean.

signed char sums[3];
int fibs[] = { 1, 1, 2, 3, 5, 8 };
char msg[] = "hello";
char msg2[] = "hello"z;
bool bs[];

Arrays of arrays are not supported.

2.2.4 Constants

Constants can be declared by using the const keyword.

const char SIZE = 42;

The compiler is quite good at detecting constants automatically, so it is not strictly necessary to declare any constants. However declaring a constant can help make the code more readable and will generate an error if any code tries to modify the value.

When an array is declared constant only the pointer to the array is constant. The contents of the array can still be modified.

2.2.5 No Floating Point Types

KickC has no floating point types, as the 6502 processor has no instructions for handling these.

2.3 Expressions

Expressions in KickC consist of operands and operators. Operands are either data type literals or names of variables or constants. All well-known expression operators from C and similar languages also exist in KickC. An example of an expression is (bits & $80) != 0.

2.3.1 Arithmetic Operators

The arithmetic operators support performing simple numeric calculations:

  • a + b: Addition
  • a - b: Subtraction
  • - a: Negation
  • + a: Positive
  • a * b: Multiplication
  • a / b: Division
  • a % b: Modulo

Multiplication, division, and remainder are allowed, however, there is limited run-time support for these operators as the 6502 has no instructions supporting them. Any multiplication/division/remainder that is a part of a calculation of a constant value is allowed. There is also runtime-support for unsigned multiplication of a variable by a constant.

2.3.2 Bitwise Operators

The bitwise operators operate on the individual bits of the numeric operands:

  • a & b: Bitwise and
  • a | b: Bitwise or
  • a ^ b: Bitwise exclusive or
  • ~ a: Bitwise not
  • a << n: Bitwise shift left n bits
  • a >> n: Bitwise shift right n bits

2.3.3 Relational Operators

The relational operators compare values and have a boolean value result:

  • a == b: Equal to
  • a != b: Not equal to
  • a < b: Less than
  • a <= b: Less than or equal to
  • a > b: Greater than
  • a >= b: Greater than or equal to

2.3.4 Logical Operators

The logical operators operate on boolean operands and have a boolean result:

  • a && b: Logical and
  • a || b: Logical or
  • ! a: Logical not

When && and || are used in if, while, do-while or similar statements they are short circuit-evaluated meaning that if a evaluates to true in if(a || b) { … } then b is never evaluated. Similarly if a evaluates to false in if(a && b) { … } then b is never evaluated.

2.3.5 Conditional Operator

The conditional operator is also called ternary operator because it is the only operator using three operands. It is used to choose between two different values.

a ? b : c Conditional

It evaluates the first operand, which should be boolean. If the first operand evaluates to true it returns the value of the second operator. If the first operand evaluates to true it returns the value of the third operator. It uses short-circuit-evaluation meaning that only one of the two last operands are evaluated depending on the value of the first operand.

In this example d will be set to the value of c if c is positive and to -c if c is negative.

char d = c>0 ? c : -c; 

2.3.6 Comma Operator

The comma operator can be used to evaluate multiple values in place of an expression. It evaluates the first operand and discards the results. Then it evaluates the second operand and returns the results.

a , b Comma

Using it can produce code that is hard to read, however it can come to good use in for()-statements to increment multiple variables. Here an example of a for()-loop with two loop variables i and j.

for(unsigned char i=0, j=0; i<32; i++, j+=2) {  }

2.3.7 Assignment Operators

Assigning a value to a variable is also an expression operator that returns the value assigned. This means that assignments can be nested inside expressions, and that they can be chained if multiple variables should be assigned the same value like a = b = 0. An assignment of course has side effects, as it modifies the assigned variable.

Compound assignment operators, such as a += b is a convenient shorthand for updating the value of a variable. It works exactly like a = a + b.

a = b	Assignment
a += b	Addition assignment
a -= b	Subtract assignment
a *= b	Multiply assignment
a /= b	Divide assignment
a %= b	Modulo assignment
a <<=b	Left shift assignment
a >>=b	Right shift assignment
a &= b	Bitwise and assignment
a |= b	Bitwise or assignment
a ^= b	Bitwise exclusive or assignment

2.3.8 Increment/decrement Operators

The pre-increment/decrement and post-increment/decrement operators is a convenient way of incrementing/decrementing the value of a numeric variable just before or just after the value is used in an expression.

  • ++a: Pre-increment
  • --a: Pre-decrement
  • a++: Post-increment
  • a--: Post-decrement

Pre-incrementing works just like incrementing the value of c before the statement, meaning that a = b + ++c; is the same as c += 1; a = b + c; Similarly post-incrementing works just like incrementing the value after the statement, meaning that a = b + c--; is the same as a = b + c; c -= 1;.

2.3.9 Pointer and Array Operators

The two basic pointer operators are the &a address-of operator, which creates a pointer to a variable and the *a pointer dereference operator, which supports reading and writing the value pointed to by the pointer.

The array indexing operator a[b], which supports reading and writing of array elements, is in fact also a pointer operator. Because an array variable is actually a pointer to the start of the array the array dereference operator a[b], which is actually shorthand for *(a+b).

  • &a: Address of
  • *a: Pointer dereference
  • a[b]: Array indexing

2.3.10 Low/High Operators

An extension in KickC is the inclusion of operators that allow addressing the low/high byte of a word, and the low/high words of a dword. These are well known from 6502 assemblers.

The low-operator <a addresses the low-byte of the word a, or the low-word if a is a dword. Similarly the high-operator >a addresses the high-byte of a word or high-word of a dword.

The low/high operator can also be used on the left side of assignments to modify only the low/high byte of a word or low/high word of a dword.

  • <a: Low part of
  • >a: High part of

The following example sets the low part of the word in a to 0. If a was $0428 before the assignment it will be $0400 afterwards.

<a = 0; 

2.3.11 Function Calls

Function that returns a value can be called as part of an expression. Functions are called by the normal parenthesis-syntax with parameter values separated by commas. How to create a function is described in the section Functions.

f(a,b): Function Call

2.3.12 Automatic Type Conversion and Type Casting

KickC handles automatic type conversions differently than standard C. Where standard C converts all small integers to int before evaluating expressions, KickC supports evaluating operators for all types and only performs conversions if they are strictly necessary. Limiting the number of automatic type conversions helps creating better optimized 6502 assembler. In many cases KickC can even handle operators for two values of different type directly. For instance adding a byte to a word can be done in more optimal 6502 code than first converting the byte to a word and then adding the two words. In practice, the difference to standard C rarely has any consequences.

Like standard C, KickC will ensure automatic type conversion (if necessary) as long as the type of one value can contain all values of the type of the other. For instance an unsigned char can be automatically converted to a signed int as signed ints can hold all possible unsigned char values. An unsigned char cannot be automatically converted to a signed char since a signed char cannot hold all possible unsigned char values.

The automatic conversions that KickC can perform for each type are described in following sections.

Type Can be automatically converted to
unsigned char unsigned int, signed int, unsigned long, signed long
signed char signed long, signed long
unsigned int unsigned long, signed long
signed int signed long
unsigned long -
signed long -

The cast operator can be used to perform explicit type conversion. Casting also allows conversions to a type that cannot hold all possible values, and where some information may be lost in the conversion, for example casting a signed char to an unsigned char.

(type)a // Casting

For sub-expressions containing only constants, the KickC compiler tries to infer the type of the sub-expression. This is done by performing the calculation and then checking which types can hold the calculated value. For instance, the calculation $4000/$80 is inferred to match any integer type except signed char (since a signed char cannot hold 128). This also differs from standard C, where all constant integer numbers are ints unless specified otherwise.

2.3.13 Operator Precedence and Parenthesis

Operators precedence decides which operators are applied first when multiple operators are combined in an expression. For instance multiplication is performed before addition in a*b+c. In KickC operators generally have the same precedence as in standard C.

Precedence rules can be overridden by using explicit parentheses in expressions. For instance to perform addition before multiplication (a+b)*c.

( a ): Parenthesis

The following table shows KickC operator precedences. Operators at the top of the table binds most tightly.

Precedence Operators Associativity
1 a++ a-- f( ) a[b] Left-to-right
2 ++a --a +a ~a !a (t)a *a &a Right-to-left
3 a*b a/b a%b Left-to-right
4 a+b a-b Left-to-right
5 a<<b a>>b Left-to-right
6 <a >a Left-to-right
7 a<b a<=b a>b a>=b Left-to-right
8 a==b a!=b Left-to-right
9 a&b Left-to-right
10 a^b Left-to-right
11 a|b Left-to-right
12 a&&b Left-to-right
13 a||b Left-to-right
14 a?b:c Right-to-left
15 a=b a+=b a-=b a*=b a/=b a%=b a<<=b a>>=b a&=b a^=b a|=b Right-to-left
16 a,b Left-to-right

2.4 Statements

The statements of a KickC program control the flow of execution. KickC supports most statements supported by standard C.

Statements are separated by semicolons stmt; stmt; and can be grouped together in blocks using curly braces { stmt; stmt; }.

2.4.1 Expressions and Assignments

All expressions are valid statements.There are two typical ways of using expressions as statements. The first is an assignment, which is an expression, modifying the value of a variable.

a += 2;

The second is calling a function that has a side effect.

print(hello);

2.4.2 If

The body of an if-statement is only executed if the condition is true. The if-statement can have an else-body, that is executed if the condition is not true.

The following if-statement prints “even” to the screen if a is even.

if((a&1)==0) { print(even); } 

The following if-statement increases b if a is less than 10 and decreases.

2.4.3 Switch

The switch-statement is used to choose between a number of cases. It is usable when you would otherwise use several consecutive if-statements. Switch works by examining a single expression and comparing the value to a number of cases labeled with a constant value. The execution starts at the case where the label matches the value. Execution then continues forward through the statements of the following cases. A break-statement can be used to break out of the switch. At the end, a special default-case can be used for catching values that were not matched by any other case.

The following switch-statement examines the char c and chooses what to do: if c is a zero, it exits the function; if c is a TAB or newline, it prints a space - and for all other values of c, it prints c itself.

switch(c) {
case 0:
	return;
case '\t':
case '\n':
	print(' ');
	break;
default:
	print(c);
}

Notice how the TAB case is empty, but since switch continues to execute statements from following cases, the execution will start at the print(' '). The return-statement in case 0 also breaks the execution of the switch.

Switch has the following syntax:

switch(expr) {
case const1:
  body1
case const2:
  body2

default:
  body3
}

Where expr is an expression giving the value to examine and const1, const2, … are expressions that must evaluate to constants. Finally, body1, body2, … are sequences of statements.

2.4.4 While

The body of a while-loop is executed repeatedly as long as the condition is still true. The while-loop is executed by first evaluating the condition. If the condition is true, the body is executed, and the loop starts over. If the condition is not true, execution continues after the loop. If the condition is not true the first time the loop is encountered, then the body is never executed.

The following while-loop prints i dots on the screen, while counting i down to zero.

while(i!=0) { print(.); i--; }

2.4.5 Do-While

The do-while-loop is very similar to the while-loop. The body of a do-while-loop is executed repeatedly as long as the condition is true. In the do-while-loop, the body is executed first, and then the condition is evaluated. If the condition is true, the loop starts over. If the condition is not true, execution continues after the loop. The body of the loop is always executed at least once.

The following do-while-loop keeps scanning the keyboard until the space key is pressed.

do {
  keyboard_event_scan();  
} while (keyboard_event_get()!=KEY_SPACE)

2.4.6 For

The for-loop is a convenient way of creating a loop, where a loop variable is initialized, the body is executed, the loop variable is incremented, and finally the condition is evaluated to determine whether to repeat the loop again.

A for-loop has the following syntax:

for(init; condition; increment) { body }

and is equivalent to the following KickC code:

init;
while(condition) {
  body;
  increment;
} 

The condition is evaluated before the body and increment, enabling for-loops where the body and increment are never executed.

KickC has an additional convenience syntax for creating simple for-loops that loop over an integer range. The following for-loop executes the body 128 times with i having values 0,1,...,127:

for(char i : 0..127) { body }    

And is equivalent to:

for(char i=0; i!=127+1; i++) { body } 

This convenience syntax only accepts constants or expressions evaluating to constants as the ends of the integer interval. It can loop both backward and forwards.

2.4.7 Break and Continue

The break statement terminates a loop or a switch, whereas a continue statement forces the next iteration of a loop. These statements are very useful when creating complex loop logic.

The following loop prints a string on the screen, skipping all spaces by using the continue statement. When it encounters a zero char in the string, it stops printing using the break statement.

char* screen = $400;
char str[] = "hello brave new world!";
for( char i: 0..255) {
	if(str[i]==0) break;
	if(str[i]==' ') continue;
	*screen++ = str[i];
}

2.5 Functions

Functions are named pieces of code that can be reused by calling them and passing different parameters.

2.5.1 Calling functions

In the following code a function called max is called 3 times. The max-function takes 2 char parameters and returns the maximal value of the two. After this code m will have the value 31 and n will have the value 47.

char m = max(31, 9 );
char n = max(m, max(47, 7));  

In general, a function is called using the syntax: name(param1, param2, …)

2.5.2 Creating functions

Functions are created by adding function declarations to your program. The following is a declaration of the max-function used above. It expects two char value parameters, finds the largest one and returns it.

char max(char a, char b) {
  if(a>b) {
    return a;
  } else {
    return b;
  }
}

In general a function declaration has the following syntax: return-type name(param-type1 param-name1, param-type1 param-name1, ...) { body }

If a function does not return a value it must declare the return type as void.

The parameter declaration inside the parenthesis describes how many parameters must be passed when calling the function, the types of the parameters to be passed and names the parameters have inside the functions body code. The max-function above takes 2 char parameters, named a and b.

The function body is the code executed when calling the function. In the body code the declared parameters can be used as variables and will have the values passed by the call.

The special statement return is used to return a value to the caller. The return statement exits the function immediately.

The following statement exits the function and returns the sum of values a and b.

return a+b;

2.5.3 The main() function

All KickC programs must have exactly one function called main. The main function is the starting point of the program. In KickC the main() function takes no parameters and returns no value.

The following is a very simple KickC program with a main function that turns the screen background color black and exits.

#include <c64.h>

void main() {
	*BGCOL = BLACK;
}

When compiling, the main-function generates a C64 BASIC program containing a single SYS command which starts the execution of the compiled KickC program.

The following is the KickAssembler code resulting from compiling the KickC program above. BasicUpstart is KickAssemblers way of creating a BASIC-program with a single SYS command.

.pc = $801 "Basic"
:BasicUpstart(main)
.pc = $80d "Program"
  .label BGCOL = $d021
  .const BLACK = 0
main: {
    lda #BLACK
    sta BGCOL
    rts
}

2.6 Comments

Two types of comments are supported in KickC. Anything inside a comment has no effect on the generated code.

Block comments, started with /* and ended with */ can span multiple lines. They can sometimes be useful for commenting out large parts of programs.

/* A multi-line
block comment */

Single line comments are started with double-slash // and ends at the next newline

// a single-line comment

2.7 Splitting code into multiple files

Like in C you can split code into multiple files by creating header-files (extension .h) and code-files (extension .c). The header-files typically contain only the interface of the module, ie. function declarations without body, extern variable declarations and #defines.

For one C-file to be able to use the functions/variables from another file you can include the header using

#include “other.h”

When including files the compiler first searches through the current folder where the file that has the #include statement is located, then it searches each library folder added by the -I / -includedir option to the compiler.

Include can be told to look in subfolders by prefixing the filename with a slash-separated path. Here is an example where graphics code files are located in a subfolder:

#include “graphics/character.h”

Using #include <file.h> instead of #include “file.h” will cause the compiler to not look for the file in the current directory, only looking in the search folders. This is useful when including libraries.

Unlike a normal C-compiler KickC does not support compiling each C-file individually and linking them later. Instead KickC insists on compiling all needed C-files in one go to be able to optimize better.

When you include a header-file using #include the compiler will try to automatically find the matching C-file. This is done by looking for a file with the same name, but extension .c in the current folder and each library folder added by the -L / -libdir option to the compiler.

You can also choose to pass multiple C-files directly to the compiler.

kickc.sh main.c other.c

KickC also differs from standard C by automatically ensuring that any included file is only included once by automatically keeping track of which files have been included. This, combined with compiling all files in one go, means that with KickC you can skip the header-files and include C-files directly instead. This may be more convenient for you if you do not plan to port your code to other C-compilers.

You can also choose to pass multiple C-files directly to the compiler by using the following command:

kickc.sh main.c other.c

KickC also differs from standard C by automatically ensuring that any included file is only included once by automatically keeping track of which files have been included. This, combined with compiling all files in one go, means that with KickC you can skip the header-files and include C-files directly instead. This may be more convenient for you if you do not plan to port your code to other C-compilers.

I provided two variations for the code block, one with inline code and one with a separate code block. You can choose the one that suits your formatting preferences.

2.8 Variables Directives

KickC has a number of directives that can be used for controlling how a variable works. Variable directives can be added before or after the type of the variable.

2.8.1 Const

The const directive is used to declare that a variable cannot be modified by the program. The declaration must also contain an assignment, and the compiler will issue an error if the variable is assigned anywhere else.

const char SPRITES = 8;

For pointers, the const directive can be used to declare either a constant pointer to a value or a pointer to a constant value. The choice is determined by the location of the const keyword relative to the *.

  • A constant pointer to a value means that the pointer cannot be modified, but the value it points to can be modified. It is declared by placing the const keyword after the *.

    char * const SCREEN = 0x0400;
    
  • A pointer to a constant value means that the program is not allowed to modify whatever the pointer points to, but the pointer itself can be modified. It is declared by placing the const keyword before the *.

    const char * ROM = 0xA000;
    

2.8.2 Register

The register directive is used to instruct the compiler to optimize a variable, putting it into a CPU-register if possible. The compiler is quite good at optimizing register usage, so using this directive should not be needed.

char register i;

It is also possible to add a specific register in parenthesis to force the variable into a specific CPU-register. Using this directive can cause the compiler to fail if it is impossible to compile the program with the variable assigned to the register.

char register(X) i;

2.8.3 Align

The __align directive is used to control the placement of arrays and strings in memory. For instance, __align(0x40) will ensure that the memory address where the data is placed is a multiple of 0x40 bytes. This can be useful when trying to optimize the performance of your program.

char __align(0x100) sine[0x100];

2.8.4 Address

The __address() directive is used to place a variable at a specific address in memory. For instance, the following will place the integer variable at address 250 (on zeropage).

__address(250) int num;

You can also use the directive for placing arrays at specific places in memory. This is quite useful for handling graphics data that must often be placed at specific addresses. The following places the array SPRITES at the memory location 0x2000.

__address(0x2000) SPRITES[64*10];

For graphics, it can be especially useful to use inline KickAssembler for initializing the data array directly from a graphics file in a modern file format. See the section "Inline KickAssembler data array initialization."

Please note that only global variable arrays can be placed at specific memory locations. __address() is not supported for local array variables.

2.8.5 Volatile

The volatile directive tells the compiler that the value of the variable might change at any time. The volatile keyword must be used for variables that are shared between code running "simultaneously." An example is when coding with interrupts (see the interrupt directive).

The directive prevents the compiler from using all optimizations, where it assumes it can guess the value of the variable from the surrounding code. It will also ensure that the compiler does not produce code where the value of the variable is held in a register.

volatile char sprite_ypos;

For pointers, the location of the volatile keyword relative to the * is used to distinguish between a volatile pointer and a pointer to volatile data. See const for a more thorough explanation.

2.8.6 Extern

The extern keyword on a variable specifies that this variable is defined somewhere else. The extern keyword is typically used in header-files when a variable defined in the C-file should be usable by other C-files.

extern char cursor_onoff;

2.8.7 Export

The export directive tells the compiler that the value of a data-variable must be included in the resulting ASM even if it is never used anywhere and would normally be deleted by the optimizer. The export keyword is only usable for global variables containing data, typically arrays.

The directive prevents the compiler from deleting the variable during optimization.

export char MESSAGE[] = "Hello World!";

2.9 Function Directives

KickC also has a few directives that instruct the compiler to treat functions in a specific way.

2.9.1 Inline

The inline function directive instructs the compiler to inline the whole function body everywhere the function is called. This can be used for optimizing your code since it allows the compiler to optimize the code of each function call independently, for instance by identifying constants in each call. It also saves the CPU cycles normally needed to call the function and return from it. The trade-off is that your program will compile into more bytes of code.

inline char sum(char a, char b) {
    return a + b;
}

2.9.2 Interrupt

The __interrupt function directive is used for creating interrupt handler functions.

__interrupt void irq() {
    *BGCOL = WHITE;
    *BGCOL = BLACK;
}

The directive can be used with parentheses to specify a specific type of interrupt handler function. Different interrupt handler function types will generate slightly different code for entering/exiting the interrupt. For instance, __interrupt(rom_sys_c64) specifies that the C64 interrupt should exit by calling the kernel standard interrupt service routine at 0xea31, which scans the keyboard and more.

If you do not specify a type in parentheses, then the default interrupt type for the platform is used.

The following are the supported interrupt types:

  • rom_min_c64: C64 Interrupt served by the kernel called through 0x0314-5. Will exit through the kernel using 0xea81, which restores the registers and exits. (Default for C64)
  • rom_sys_c64: C64 Interrupt served by the kernel called through 0x0314-5. Will exit through the kernel using 0xea31, which runs the normal kernel service routine that includes checking and handling keyboard input.
  • rom_min_vic20: VIC20 Interrupt served by the kernel called through 0x0314-5. Will exit through the kernel using 0xeb18, which restores the registers and exits. (Default for VIC20)
  • rom_sys_vic20: VIC20 Interrupt served by the kernel called through 0x0314-5. Will exit through the kernel using 0xeabf, which runs the normal kernel service routine that includes checking and handling keyboard input.
  • rom_min_plus4: Plus/4 Interrupt served by the kernel called through 0x0314-5. Will exit through the kernel using 0xfcc3, which restores the registers and exits. (Default for Plus/4)
  • rom_sys_plus4: Plus/4 Interrupt served by the kernel called through 0x0314-5. Will exit through the kernel using 0xce0e, which runs the normal kernel service routine that includes checking and handling keyboard input.
  • rom_min_cx16: CX16 Interrupt served by the kernel called through 0x0314-5. Will exit through the kernel using 0xe049, which restores the registers and exits. (Default for CX16)
  • rom_sys_cx16: CX16 Interrupt served by the kernel called through 0x0314-5. Will exit through the kernel using 0xe034, which runs the normal kernel service routine that includes checking and handling keyboard input.
  • hardware_all: Interrupt served directly from hardware through 0xfffe-f or 0xfffa-b. Will exit through RTI and will save/restore all registers. (Default for Atari 2600, Atari XL, and NES)
  • hardware_clobber: Interrupt served directly from hardware through 0xfffe-f or 0xfffa-b. Will exit through RTI and will save necessary registers based on a clobber analysis of the interrupt handler code.
  • hardware_none: Interrupt served directly from hardware through 0xfffe-f or 0xfffa-b. Will exit through RTI and will save/restore NO registers.

The hardware-interrupt types can be used on all platforms, since they do not rely on any code in ROM.

In general, interrupts are set up by assigning a pointer to an interrupt handler function to one of the interrupt vectors placed at a specific address in memory. Below is a C64 example of setting up the kernel IRQ vector (at $314 in memory) to run the irq() function declared above. When setting up interrupts, it is good practice to surround the code with the SEI/CLI instructions to prevent any interrupt from occurring during the setup itself.

#include <c64.h>
#include <6502.h>

void main() {
    SEI();
    *KERNEL_IRQ = &irq;
    CLI();
}

If your interrupt code needs to utilize global variables to communicate with other parts of the program or to store state between interrupt calls, these variables should be declared as volatile.

2.9.3 Inline Assembler Code

Programs can include inline assembler code inside a function body. This can, for instance, be useful for interfacing with machine code such as the BASIC/KERNAL or for modifying processor flags (such as the interrupt flag or decimal flag).

Inline assembler is created using the asm statement with a body containing the assembler code. The following is an example of setting the processor interrupt flag.

asm {
  sei
}

The assembler language usable within the curly braces is pretty limited standard syntax 6502/65C02/65CE02/45GS02 assembler using the same syntax as KickAssembler (and most classical 6502 assemblers). The following is supported:

  • All 6502/6510 instructions and addressing modes
  • Immediate lda #%10101010
  • Absolute eor 1024
  • Zeropage rol 2
  • Relative bne nxt
  • Absolute indexed X adc $2000,x
  • Absolute indexed Y cmp sintab,y
  • Zeropage indexed X sbc 2,x
  • Zeropage indexed Y stx $fe,y
  • Zeropage indexed indirect X lda ($20,x)
  • Zeropage indirect indexed Y ora (14),y
  • Indirect jmp ($1000)
  • Implied (no operand) tax

The 65CE02 instructions and addressing modes:

  • Indirect Zeropage Indexed by Z lda ($12),z
  • Stack Pointer Indirect Indexed sta ($12,sp),y
  • Immediate Word phw #$1234
  • Long Relative lbra #$1234

The 45GS02 instructions and addressing modes:

  • 32-bit Indirect Zeropage Z lda (($12)),z
  • 32-bit Indirect Zeropage stq (($12))

Labels:

  • Normal labels next:
  • Multi labels !next:

Data:

  • Bytes .byte $10, $20

The parameters for instructions and the data bytes can be written as expressions supporting:

  • Literal numbers in decimal, binary, or hexadecimal using the same syntax as KickC literal numbers, e.g., 1024, $3fff, %10101010
  • Literal characters, e.g., 'q'
  • Constant variables declared in the C-code, e.g., SCREEN
  • Labels declared within the assembler code
  • Math operators + - * / < > << >>
  • Parentheses using [ and ] to avoid the assembler interpreting them as an indirect addressing mode.

The ability to reference constant variables declared outside the assembler code allows the inline assembler to interact with data in the C-part of the program. The following is an example referencing the constant variable SCREEN.

const char* SCREEN = $400;

void main() { 
	asm {
		lda #'c'
		sta SCREEN+40
	}
}

There is also support for referencing labels declared inside inline ASM in other functions using the special . operator (e.g., clearscreen.fillchar). This makes it possible for ASM in one function to modify the code of ASM inside other scopes.

The KickC compiler understands the inline assembler code and attempts to optimize it during the compilation process. For instance, it can analyze which registers are clobbered by the inline assembler, and optimize surrounding KickC-code register usage.

If your inline ASM contains a JSR call, the compiler assumes that all registers are clobbered. However, it is possible to add a clobbers directive in parenthesis specifying exactly which registers are clobbered by your inline assembler code. Here is an example of inline assembler where the directive is used to specify that the JSR only clobbers the A- and X-registers.

void playMusic() {
	asm(clobbers "AX") {
		jsr $1000
	}
}

Since the compiler understands the inline assembler, it will also modify the assembler code if this leads to faster execution, for instance removing an immediate load-instruction that loads a value that the register is already guaranteed to contain.

2.9.4 Inline KickAssembler Code

The inline assembler code described above can be very useful, but only supports rudimentary assembler features. The limitations allow the compiler to understand the assembler code and include it in optimizations.

If you need advanced assembler features in your code such as macros, loops, or importing binary files or images, it is possible to include inline KickAssembler code in your KickC program using the kickasm statement. The body of the kickasm statement must be enclosed in double curly braces and is passed directly to KickAssembler. The KickC compiler does not make any attempt to parse or understand the KickAssembler code. All advanced KickAssembler features are described in the manual here.

The following is an example of inline KickAssembler code creating assembler for really fast screen clearing (1000 STA operations with no looping).

void clearscreen() {
  kickasm(uses screen) {{
    lda #0
    .for (var i = 0; i < 1000; i++) {
      sta screen+i
    }
  }}
}

Inline KickAssembler can reference constant variables declared in the surrounding C-code. To ensure that the KickC compiler knows that the inline KickAssembler uses a constant, you should add a uses directive in parentheses. This will ensure that KickC knows that the symbol is used and, for instance, prevent the compiler from removing the symbol entirely if it is not used anywhere else.

2.9.5 Inline KickAssembler data array initialization

Inline KickAssembler can also be used to initialize data arrays. Here it allows utilizing KickAssembler's powerful macro language. The following is an example of a sine table:

char sintab[0x100] = kickasm {{
    .fill $100, 127.5 + 127.5*sin(i*2*PI/256)
}};

To specify exactly where the resulting data bytes end up in memory, the __address() directive can be used. Here's an example of generating a table with sine values at a specific location:

__address(0x1000) char sintab[] = kickasm {{
    .fill $100, 127.5 + 127.5*sin(toRadians(i*360/256))
}};

Please note that the __address() directive is only allowed for global variables.

It is also possible to use inline KickAssembler for loading pictures, music, or other binary files and generating data bytes from these. When loading binary files in the inline KickAssembler code, it is necessary to inform the KickC compiler using a resource directive within parentheses. This is needed because KickC may have to copy the used resource files to the output directory where the compiled assembler code is written. Here's an example of including a sprite from a PNG image file and placing it at a specific memory address:

__address(0xc00) char SPRITE[] = kickasm(resource "balloon.png") {{
    .var pic = LoadPicture("balloon.png", List().add($000000, $ffffff))
    .for (var y=0; y<21; y++)
        .for (var x=0; x<3; x++)
            .byte pic.getSinglecolorByte(x, y)
}};

2.10 Controlling the Compilation

KickC offers a number of ways for you to tell the compiler how to perform the compilation. This can either be done through #pragma commands in the source code or through command line options. The #pragma commands can offer the advantage that your compilation options are documented directly in the source code.

2.10.1 Target Platform

You can specify which target platform the compiler should compile for by using #pragma target(name) or the command line options -t name or -p name.

The target platform collects a linker file and a number of other compilation options into a single convenient choice that makes it easy to create binary files for a specific platform. The target platform also specifies the default emulator used to execute the binaries.

The target platforms are specified through the tgt-files in the target folder in the KickC installation. You can add your own target platform by creating a new tgt-file and an ld-file (a linker specification file).

The KickC-compiler includes the following target platforms out-of-the-box:

  • Commodore VIC 20
  • Commodore 64
  • Commodore Plus/4 (also usable for Commodore 16 and Commodore 116)
  • Atari 2600
  • Atari XL/XE
  • Nintendo NES
  • MEGA65
  • Commander X16

2.10.2 Linker Specification

You can create your own linker specification describing exactly how to generate a binary file from the code and data of your program.

You can tell the compiler to use a custom linker specification file by using #pragma link("filename") or the command line options -T filename.

The linker specification file uses the KickAssembler segment format, as described in chapter 10 and 11 of the KickAssembler manual. See KickAssembler Manual.

KickC expands the following strings in the linker script before passing it on to KickAssembler:

  • %E: Expands to the name of the entry point in the KickC program (i.e., the label of the code that starts the program, typically main or __start)
  • %P: Expands to the start address, where the generated code should be located in memory. See #pragma start_address()
  • %O: Expands to the name of the output file that the resulting binary should be saved to. The output filename is the same as the primary input file with the extension of a binary program. The binary file extension depends on the target platform or can be specified using #pragma extension.
  • %_O: Expands to a lowercase version of the output file name from %O.
  • %^O: Expands to an uppercase version of the output file name from %O.

KickC includes linker specifications for all supported target platforms. These are the files with the extension ld in the target folder. These can be used as a starting point for your own linker specification files.

Some of the included example programs demonstrate how to use custom linker files. For instance, the example program kernalload.c uses a custom linker file for generating a binary D64 disk image.

2.10.3 Code and Data Segments

By default, your code is compiled into the segment named "Code," and your data is compiled into the segment named "Data."

If you create your own linker specification file, you can also declare new named segments that the linker file then outputs into the binary file.

In your program, you can specify which named segment the code should be compiled into using #pragma code_seg(name). Similarly, you can specify which named segment the data should be compiled into using #pragma data_seg(name). These two #pragmas can be used multiple times in your source code. They take effect when encountered and are used for all following code and data until you instruct the compiler to use another named segment using the #pragma once more.

2.10.4 Target CPU

You can specify which CPU the compiler should create assembler code for by using #pragma cpu(name) or the command line option -cpu name.

The following CPUs are supported:

  • rom6502: Vanilla MOS 6502 CPU running in ROM (no illegal opcodes, no self-modifying code).
  • rom6502x: MOS 6502 CPU running in ROM (allows illegal instructions, no self-modifying code).
  • mos6502: Vanilla MOS 6502 CPU (no illegal opcodes, allows self-modifying code).
  • mos6502x: MOS 6502 CPU (allows illegal instructions, allows self-modifying code).
  • wdc65c02: WDC 65C02 CPU (More addressing modes and instructions, no self-modifying code).
  • csg65ce02: CSG 65CE02 CPU (Even more addressing modes and instructions, no self-modifying code).
  • mega45gs02: MEGA65 45GS02 CPU (Even more addressing modes and instructions, no self-modifying code).

Support for the 65CE02 and 45GS02 CPUs is achieved by using a modified version of KickAssembler. The modified KickAssembler will be retired when the KickAssembler officially adds support for these CPUs. It is available here.

2.10.5 Character Encoding

You can specify which character encoding the compiler uses to find the numerical value of a character or string by using #pragma encoding(name).

The following character encodings are supported:

  • petscii_mixed: Commodore PET Standard Code of Information Interchange mixed case.
  • petscii_upper: Commodore PET Standard Code of Information Interchange uppercase.
  • screencode_mixed: Commodore screen codes mixed case (Default for VIC20, C64, Plus4, MEGA65).
  • screencode_upper: Commodore screen codes uppercase.
  • atascii: ATARI Standard Code for Information Interchange (Default for Atari 2600).
  • screencode_atari: ATARI screen codes (Default for Atari XL).
  • ascii: American Standard Code for Information Interchange.

2.10.6 Interrupt Type

You can specify the interrupt type that is used for functions with the __interrupt directive by using #pragma interrupt(type). This instructs the compiler what interrupt type to use for a function with an __interrupt directive without a parenthesis. You can see the available interrupt types in the section "Interrupt" above.

2.10.7 Program Start Address

You can specify the program start address in memory by using #pragma start_address(number). The program start address is passed on to the linker script, which uses it to place the program in memory. The default start address for each target platform is specified in the tgt-file.

2.10.8 Zeropage Address Reservations

If you want the compiler to avoid using some parts of the zeropage because you plan to use them for other purposes, then you can specify this using #pragma zp_reserve(number, number, ...). If you want to avoid a whole range of zeropage addresses, you can specify that using the syntax number..number as one of the parameters. For example, the following instructs the compiler to not use zeropage addresses 2, 7, and 0x10-0x1f.

#pragma zp_reserve(2, 7, 0x10..0x1f)

2.10.9 Emulator

You can specify which emulator to use for running the compiled binary program by using #pragma emulator("command"). When the program is compiled with the -e command line option, KickC will run the emulator command and pass it

the generated binary program. The target platform tgt-files define the default emulator command for each platform, so you only need to use the pragma if you want to override the default.

2.10.10 Binary Output File Extension

You can override the filename extension used for binary output files by using #pragma extension("ext"). The default extension for the binary output files is specified in the target platform tgt-file.

2.10.11 Library Init Code

If you are coding a library, you can specify that a function is init-code that needs to run before the main() program is started. If you specify the following, then the compiler will automatically add a call to the function init() before calling main() to any program that uses the functions proc1, proc2, or proc3.

#pragma constructor_for(init, proc1, proc2, proc3)

If proc1, proc2, or proc3 are never used, then the init function is not called, and the KickC optimizer will delete it unless it is explicitly called elsewhere.

2.12 Comparison with Standard C99

2.12.1 Not Supported/Implemented

  • Floating point types: float, double
  • Runtime multiplication: a * b (except constants)
  • Runtime division: a / b (except powers of 2)
  • Runtime modulo: a % b (except powers of 2 - 1)
  • Union: union { char b; int w; } u;
  • Array of arrays: char baa[4][4];
  • Function pointers with parameters: void(char)*;
  • Function pointers with return values: char()*;
  • Recursive functions: char fib(byte n) { return fib(n-1) + fib(n-2); }
  • Alignof operator: char s = alignof(word);
  • Variadic functions: sum(int i1, ...) (except printf)

2.12.2 Limitations/Modifications

  • Multiplication, division, and modulo have limited support.
    • Multiplication is supported for constant values.
    • Division is supported for constant powers of 2.
    • Modulo is supported for constant powers of 2 minus 1.
    • Multiplication of a variable with an unsigned constant is supported and converted to shifts/additions.
  • There is no general runtime support for multiplication/division/modulo without using a library and a function call.
  • Arrays and strings are always statically allocated (as data bytes in the resulting assembler).
  • Alignment directive: __align($100)
  • Register directive: register(X)
  • Inline assembler: asm { SEI CLD };
  • Main function: void main() { ... }

2.12.3 Extensions

  • Forward referencing variables in the outer scope
  • Ranged for-loops: for( char i: 0.. 10) { }
  • Word operator: unsigned int w = { hi, lo };
  • Lo/hi-byte operator: char lo = <w; <w = 12;

2.12.4 Tips for Optimizing KickC Code

  • Use char/byte anywhere possible. They are faster than int/word.
  • Use unsigned types whenever possible; they are faster than signed types.
  • Use array indexing instead of incrementing pointers.
  • Use do {} while() instead of while() {} for loops that always execute at least once.
  • Booleans are not always very efficient; often, char is better.
  • Use inline functions.
  • Use (experimental) inline loops.
  • Use normal optimization techniques.
  • Put calculated results that are used multiple times into a variable instead of repeating the calculation.
  • Create arrays for lookup instead of repeating a calculation many times.
  • Unroll loops (manually).

3 Working with KickC

3.1 Controlling Output Files

Running the compiler will generate a number of output files. The directory and name of the output files all have sensible defaults and can be controlled through configuration or command-line options.

The following output files generated are:

  1. An ASM-file containing the assembler code for all the C-files. This ASM-file collects assembler code for all C-files passed to the compiler and all used functions in all included libraries.
  2. A copy of all resource-files needed to compile the ASM-file. Resource-files are defined in inline kickass and typically contain images, music, data tables, etc. Resource-files are copied to the output directory if it is not the same as the input directory.
  3. A binary program file that is the result of assembling the ASM-file. The binary output file is generated when instructing the compiler to also do the assembly by command-line options -a or -e. The contents and specific format of the binary program file are controlled by the linker specification of the target platform (or any custom linker specification). The linker specification can potentially generate multiple binary output files. In the linker specification, %O is used to expand the binary output file name.

The binary program output file name generated has the following file name components:

  • Output-path is the directory where the output files are generated. The following is a prioritized list specifying how the compiler finds this folder:
    • If the -o command line option is passed, and the passed file name contains a directory part, then this directory is used.
    • If the -odir command line option is passed, then this directory is used.
    • If the first C-file passed to the compiler contains a directory part, then this directory is used.
    • Otherwise, the current directory is used.
  • Output-basename is the file name of the output file without path and extension. The following is a prioritized list specifying how the compiler finds the output basename:
    • If the -o command line option is passed, then the basename of the passed file name is used.
    • The basename of the first C-file passed to the compiler is used.
  • Output-extension is the file extension of the output file. The following is a prioritized list specifying how the compiler finds the output extension:
    • If the -o command line option is passed, and the compiler has been instructed to assemble the program (using -a, -e, or -d), then the extension of the passed file name is used.
    • If the program contains #pragma extension(...), then the specified extension is used.
    • If the -oext command line option is passed, then this extension is used.
    • Otherwise, the extension specified in the platform target (.tgt) file for the chosen platform is used.

The ASM output file has the following file name components:

  • Output-path / output-basename . asm-extension

Asm-extension is the file extension of the ASM file. The following is a prioritized list specifying how the compiler finds the output extension:

  • If the -o command line option is passed, and the compiler has not been instructed to assemble the program (using -a, -e, or -d), then the extension of the passed file name is used.
  • Otherwise, the extension "asm" is used.

3.2 KickC Command Line Reference

kickc [-adeEhSvV] [-Ocoalesce] [-Oliverangecallpath] [-Oloophead] [-Onocache]
        [-Onoloophead] [-Onouplift] [-Si] [-Sl] [-vasmoptimize] [-vasmout]
        [-vcreate] [-vfragment] [-vliverange] [-vloop] [-vnonoptimize]
        [-voptimize] [-vparse] [-vsequence] [-vsizeinfo] [-vunroll] [-vuplift]
        [-Warraytype] [-Wfragment] [-calling=<calling>] [-cpu=<cpu>]
        [-emu=<emulator>] [-F=<fragmentDir>] [-fragment=<fragment>]
        [-o=<outputFileName>] [-odir=<outputDir>] [-oext=<outputExtension>]
        [-Ouplift=<optimizeUpliftCombinations>] [-p=<targetPlatform>]
        [-T=<linkScript>] [-var_model=<varModel>] [-D=<String=String>]...
        [-I=<includeDir>]... [-L=<libDir>]... [-P=<targetPlatformDir>]...
        [-Xassembler=<assemblerOptions>]... [<cFiles>...]

3.2.1 Description:

Compiles C source files. KickC is a C-compiler creating optimized and readable 6502 assembler code.

See KickC Repository for detailed information about KickC.

3.2.2 Parameters:

  • <cFiles>...: The C source files to compile.

3.3.3 Options:

  • -a: Assemble the output file using KickAssembler. Produces a binary file.
  • -calling=<calling>: Configure calling convention. Default is __phicall. See #pragma calling.
  • -cpu=<cpu>: The target CPU. Default is 6502 with illegal opcodes. See #pragma cpu.
  • -d: Debug the assembled binary file using C64Debugger. Implicitly assembles the output.
  • -D=<String=String>: Define macro to value (1 if no value given).
  • -e: Execute the assembled binary file using an appropriate emulator. The emulator chosen depends on the target platform.
  • -E: Only run the preprocessor. Output is sent to standard out.
  • -emu=<emulator>: Execute the assembled program file by passing it to the named emulator. Implicitly assembles the output.
  • -F, -fragmentdir=<fragmentDir>: Path to the ASM fragment folder, where the compiler looks for ASM fragments.
  • -fragment=<fragment>: Print the ASM code for a named fragment. The fragment is loaded/synthesized, and the ASM variations are written to the output.
  • -h, --help: Show this help message and exit.
  • -I, -includedir=<includeDir>: Path to an include folder, where the compiler looks for included files. This option can be repeated to add multiple include folders.
  • -L, -libdir=<libDir>: Path to a library folder, where the compiler looks for library files. This option can be repeated to add multiple library folders.
  • -o, -output=<outputFileName>: Name of the output file. By default, it is the same as the first input file with the proper extension.
  • -Ocoalesce: Optimization Option. Enables zero-page coalesce pass which limits zero-page usage significantly, but takes a lot of compile time.
  • -odir=<outputDir>: Path to the output folder, where the compiler places all generated files. By default, the folder of the output file is used.
  • -oext=<outputExtension>: Override the output file extension. The default is specified in the target platform / linker configuration.
  • -Oliverangecallpath: Optimization Option. Enables live ranges per call path optimization, which limits memory usage and code, but takes a lot of compile time.
  • -Oloophead: Optimization Option. Enabled experimental loop-head constant pass which identifies loops where the condition is constant on the first iteration.
  • -Onocache: Optimization Option. Disables the fragment synthesis cache. The cache is enabled by default.
  • -Onoloophead: Optimization Option. Disabled experimental loop-head constant pass which identifies loops where the condition is constant on the first iteration.
  • -Onouplift: Optimization Option. Disable the register uplift allocation phase. This will be much faster but produce significantly slower ASM.
  • -Ouplift=<optimizeUpliftCombinations>: Optimization Option. Number of combinations to test when uplifting variables to registers in a scope. By default, 100 combinations are tested.
  • -p, -t, -target, -platform=<targetPlatform>: The target platform. Default is C64 with BASIC upstart. See #pragma target.
  • -P, -targetdir, -platformdir=<targetPlatformDir>: Path to a target platform folder, where the compiler looks for target platform files. This option can be repeated to add multiple target platform folders.
  • -S, -Sc: Interleave comments with C source code in the generated ASM.
  • -Si: Interleave comments with intermediate language code and ASM fragment names in the generated ASM.
  • -Sl: Interleave comments with C source file name and line number in the generated ASM.
  • -T, -link=<linkScript>: Link using a linker script in KickAss segment format. See #pragma link.
  • -v: Verbose output describing the compilation process.
  • -V, --version: Print version information and exit.
  • -var_model=<varModel>: Configure variable optimization/memory area. Default is ssa_zp. See #pragma var_model.
  • -vasmoptimize: Verbosity Option. Assembler optimization.
  • -vasmout: Verbosity Option. Show KickAssembler standard output during compilation.
  • -vcreate: Verbosity Option. Creation of the Single Static Assignment Control Flow Graph.
  • -vfragment: Verbosity Option. Synthesis of Assembler fragments.
  • -vliverange: Verbosity Option. Variable Live Range Analysis.
  • -vloop: Verbosity Option. Loop Analysis.
  • -vnonoptimize: Verbosity Option. Choices not to optimize.
  • -voptimize: Verbosity Option. Control Flow Graph Optimization.
  • -vparse: Verbosity Option. File Parsing.
  • -vsequence: Verbosity Option. Sequence Plan.
  • -vsizeinfo: Verbosity Option. Compiler Data Structure Size Information.
  • -vunroll: Verbosity Option. Loop Unrolling.
  • -vuplift: Verbosity Option. Variable Register Uplift Combination Optimization.
  • -Warraytype: Warning Option. Non-standard array syntax produces a warning instead of an error.
  • -Wfragment: Warning Option. Missing fragments produce a warning instead of an error.
  • -Xassembler=<assemblerOptions>: Passes the next option to the assembler. The option should generally be quoted. This option can be repeated to pass multiple options.

4 The Compiler Architecture

The KickC compiler utilizes the following phases during compilation:

4.1 Loading and Parsing

Loads the main source file and recursively loads all imported source files. Parses the files creating a parse tree representation.

4.2 Creating Control Flow Graph and Symbol Table

Converts the parse tree to a symbol table containing all variables and functions and a control flow graph in static single assignment form (SSA form). SSA form consists of a control flow graph which models all possible execution paths. Each control flow block contains a number of statements (mostly assignments). Expressions are broken into a number of simple assignments to intermediate variables ensuring that each SSA statement only handles a single operator. In SSA form variables are also broken into multiple versions to ensure that each variable is assigned a value exactly once in the entire program. Finally transitions between blocks are handled through so called Phi-functions that describe how variable versions are mapped when execution flows from one block to another. SSA form has huge advantages in making a lot of optimizations easier to program.

4.3 Optimizing the SSA Control Flow Graph

Optimizes the control flow graph by repeatedly calling 20+ micro-passes. Each micro-pass examines the control flow graph and can perform a specific type of optimization by modifying the graph. The optimizer keeps cycling through the micro-passes until none of them can perform any more optimizations.

4.4 Control Flow Graph Analysis

Performs several analyses of the program in preparation of register allocation. This includes call graph analysis (which functions call each other), variable live range analysis (where is a variable defined, where is it used and where is it alive, meaning that it will be used at a later point) and loop analysis (which loops exist in the code, which loops nest each other, how deep is each loop).

4.5 Register Allocation

Allocates registers and memory locations to all variables. Initially in this phase the SSA variables are grouped together into groups that will benefit from having the same register allocation. The technique used for this is called PhiLifting and PhiMemCoalesce. The allocation is then essentially done by trying out a lot of different register combinations and choosing the one that generates the best assembler code (uses fewest cycles). The number of different combinations tested can be controlled by the compiler option -Ouplift=nnnn

4.6 Assembler Code Generation using Assembler Fragments

6502 Assembler code is generated by converting each SSA statement to assembler code using the chosen register allocation. Because SSA statements can only express pretty simple operations ASM generation is essentially done by having a large library of ASM fragment files containing the ASM code needed for a specific SSA statement with a specific register allocations. This library of ASM fragments is stored in the fragment folder in the compiler installation. The Fragment sub-system of the compiler loads fragments from this folder, but also uses a bunch of rules for synthesizing more advanced fragments from simpler ones. Even with the current 500+ fragment-files and 150+ synthesis-rules the compiler still occasionally runs into SSA statements it does not know how to create assembler for. If you encounter this problem you can fix it yourself by adding the right fragment-file in the fragments-folder. The fragment sub-system is described in more detail below to help you do this in case you need to.

4.7 Assembler Code Optimization

Finally the compiler performs optimization of the generated assembler code. This is also done by repeatedly calling a bunch of micro-passes that each knows how to perform a simple Assembler Code Optimization. The optimizer keeps cycling through the micro-passes until none of them can perform any more optimizations. The assembler optimizations include eliminating redundant register loads (eg.LDA #0 when A is already zero), replacing double jumps with jumps straight to the destination, removing unused labels, eliminating etc.

5 Assembler Fragment Sub System

5.1 Fragment File Name

The format for values in a fragment name is always 5 letters, which describe the type and the value and where it is stored. An example is vbuz1, which is an unsigned byte value stored in zero-page memory.

The first 3 letters describe the type, and the last 2 describe the storage.

Integer types are described by the following:

  1. v value / p pointer / q pointer to pointer
  2. b byte / w word / d dword
  3. u unsigned / s signed

There are also a few special types with the following 3-letter names:

  • vss struct value
  • pss pointer to struct value
  • vbo boolean value
  • pbo pointer to boolean value

The storage of the value is described by 2 letters:

  • aa A-register
  • xx X-register
  • yy Y-register
  • z1 zeropage-memory {z1}
  • m1 main-memory {m1}
  • c1 constant {c1}

When {c1} is used for values, it is an immediate value, e.g., in vbuc1, {c1} is a constant unsigned byte value.

When c1 is used for pointers, it is an address in main memory, e.g., in pbsc1, {c1} is a constant pointer to a signed byte. This means that {c1} is effectively an address in main memory.

vwuz1_gt_vbsc1_then_la1, for example, means [if] variable word unsigned zero-page value {z1} is greater than variable byte signed constant {c1}, then [goto] label {la1}.

Fragments can use $ff as temporary storage (and $fe if 2 addresses are needed).

If you are wondering how a specific fragment looks, you can ask the compiler using the -fragment flag. The following command will show all the different variations of assembler the compiler can use when needing to assign an unsigned byte in a zeropage variable {z1} to the value found in a table of unsigned bytes {c1} indexed by another unsigned byte variable on zeropage {z2}.

kickc.sh -vfragment -fragment vbuz1=pbuc1_derefidx_vbuz2 

5.2 Adding Missing Fragments

The following is my recipe for adding a fragment. Say, for instance, that the fragment vbuz1=pbuz2_derefidx_vbuc1_minus_vbuxx is missing. I start by calling the following command to see what different ways I can choose to add the fragment:

kickc.sh -vfragment -fragment vbuz1=pbuz2_derefidx_vbuc1_minus_vbuxx

The output from the command is a list of all the different fragments that the compiler could use to synthesize the fragment that is missing. If I add any of these, then the compiler can use that fragment to create the missing fragment.

  • New fragment synthesis vbuz1=pbuz2_derefidx_vbuc1_minus_vbuxx
  • New fragment synthesis vbuz1=pbuz2_derefidx_vbuc1_minus_vbuxx - sub-option vbuz1=pbuz2_derefidx_vbuc1_minus_vbuaa
  • New fragment synthesis vbuz1=pbuz2_derefidx_vbuc1_minus_vbuxx - sub-option vbuz1=pbuz2_derefidx_vbuc1_minus_vbuaa
  • New fragment synthesis vbuz1=pbuz2_derefidx_vbuc1_minus_vbuxx - sub-option vbuz1=pbuz2_derefidx_vbuc1_minus_vbuyy
  • ...

Fragment synthesis vbum1=pbuz2_derefidx_vbuc1_minus_vbuaa - No file or synthesis results! Fragment synthesis vbum1=pbuz2_derefidx_vbuc1_minus_vbuyy - No file or synthesis results! Fragment synthesis vbuz1=pbuz2_derefidx_vbuc1_minus_vbuyy - No file or synthesis results! Fragment synthesis vbuz1=pbuz2_derefidx_vbuc1_minus_vbuaa - No file or synthesis results! Fragment synthesis vbuz1=pbuz2_derefidx_vbuc1_minus_vbuxx - No file or synthesis results!

I look through the list and try to choose the most general of the fragments that I can write optimal ASM for. By general, I mean the fragment that will allow the compiler to synthesize as many fragments as possible.

My rules of thumb for choosing the most general are:

  • A fragment that works on registers (like vbuxx) is more general than fragments that work on zero-page/memory (like vbum1). The compiler is good at synthesizing memory-based fragments from register-based fragments, so adding the register-based fragment will add many more potential fragments than adding the memory-based fragment.
  • Fragments that use main-memory (like vbum1) are more general than fragments that use zero-page memory (like vbuz1). The compiler can convert main-memory access to zero-page access, so adding that will effectively also add the zero-page fragment. Only if the zero-page addressing is indirect, I choose to make a zero-page-based fragment (since 6502 does not allow indirect addressing in main-memory).
  • Fragments that are shorter are better than longer fragments. Sometimes the compiler can synthesize a whole sub-part of the fragment, and you only need to add a short fragment.

For the example above, I would choose to add the fragment vbuaa=pbuz1_derefidx_vbuyy_minus_vbuaa.asm with the following body:

eor #$ff
sec
adc ({z1}),y

After adding it, I check that it can be used to create the fragment that was missing in the first place, and that the ASM looks good by calling the command again:

kickc.sh -vfragment -fragment vbuz1=pbuz2_derefidx_vbuc1_minus_vbuxx

This results in the same list as before but now also has information about how the fragment was synthesized and the actual ASM.

  • ...
  • Fragment synthesis vbuz1=pbuz2_derefidx_vbuc1_minus_vbuaa - Successfully synthesized from vbuz1=pbuz2_derefidx_vbuxx_minus_vbuaa
  • synthesized vbuz1=pbuz2_derefidx_vbuc1_minus_vbuxx < vbum1=pbuz2_derefidx_vbuc1_minus_vbuxx < vbuaa=pbuz1_derefidx_vbuc1_minus_vbuxx < vbuaa=pbuz1_derefidx_vbuyy_minus_vbuxx < vbuaa=pbuz1_derefidx_vbuyy_minus_vbuaa - clobber:A Y cycles:16.5 ldy #{c1} txa eor #$ff sec adc ({z2}),y sta {z1}

5.3 Handling Clobbered Registers

You do not need to restore any register values in fragments. In fact, part of the optimizer's strength is that it keeps track of which registers are modified by which fragments.

The compiler analyzes the ASM in each fragment and determines which registers are clobbered. When allocating variables to registers, it avoids any allocation where a fragment clobbers a register holding a variable value that is needed later in the code. So it will avoid holding a value in A that is needed after any fragment that clobbers A - and will instead look at different options (X, Y, or on zero page).

This produces much better ASM than each fragment restoring register values since it allows the compiler flexibility in choosing the register/zeropage allocation that minimizes the number of cycles the code consumes.

This is further improved by the compiler treating each assignment to a variable as a separate variable - meaning it often ends up choosing to hold much-used variables in different registers or on zero page for different parts of the code.

Overall KickC has the following compile process:

  1. Loading and Parsing
  2. Creating Control Flow Graph and Symbol Table
  3. Optimizing the SSA Control Flow Graph
  4. Control Flow Graph Analysis & Register Allocation
  5. Assembler Code Generation using Assembler Fragments
  6. Assembler Code Optimization

The Control Flow Graph uses Single Static Assignment (where each variable is only assigned once - and where statements are mostly assignments with 1 or 2 arguments and an operator.) See Static Single Assignment Form.

The SSA of KickC has been specifically designed to inline pointer derefs or indexed pointer derefs - because this improves the ASM that it generates significantly.

The SSA Optimization optimizes the control flow graph by repeatedly calling 20+ micro-passes. This is where constants are propagated, unused code is removed and so forth.

Register Allocation decides which variables to put into registers (A/X/Y) and which to store on zeropage. This is done based on variable range live range analysis, using PHI-lifting and register clobber analysis - plus testing a lot of combinations to find the solution using the fewest CPU cycles. See Register Allocation and Assignment.

The SSA-statements are then turned into ASM using the fragment system. Each SSA-statement combined with a specific register allocation becomes a fragment.

An example is the fragment:

vbuz1 = vbuz1_plus_pbuc1_derefidx_vbuxx

which matches an SSA-statement like:

digit#1 = digit#3 + *(UTOA10_VAL#0 + i#2)

where digit#1 and digit#3 are unsigned bytes allocated to the same zeropage address (which is possible because digit#3 is never used again after this assignment so digit#1 can overwrite it), UTOA10_VAL#0 is a table of unsigned byte values stored in memory at a constant address, and i#2 is an index into the table allocated to the X-register.

The fragment sub-system has two parts. First around 900 files with specific fragment ASM-code. Second, a fragment synthesizer that can make more complex fragments from simpler fragments.

The fragment above is actually synthesized from vbuaa = vbuaa_plus_vbuz1 (which is just adding a value on zeropage to the A-register) using different synthesis rules.

synthesized vbuz1 = vbuz1_plus_pbuc1_derefidx_vbuxx < vbuz1 = pbuc1_derefidx_vbuxx_plus_vbuz1 < vbuaa = pbuc1_derefidx_vbuxx_plus_vbuz1 < vbuaa = vbuz1_plus_pbuc1_derefidx_vbuxx < vbuaa = vbuz1_plus_vbuaa < vbuaa = vbuaa_plus_vbuz1 - clobber: A cycles: 12.5

lda {c1},x
clc
adc {z1}
sta {z1}

Finally, the resulting ASM is optimized using peephole optimization.